The UK’s AI Safety Institute put Claude Mythos Preview through its paces and confirmed Anthropic’s claim: the model excels at spotting security flaws. Their report, led by Drew Breunig, shows a clear link between tokens spent and vulnerabilities found. In practice, that means if you pour more money into reviews, Mythos keeps uncovering fresh exploits. Hardening a system becomes a matter of outspending potential attackers on token-driven audits. That payment-per-token dynamic flips the value equation for libraries. When you invest tokens into an open source project, every downstream user shares in the benefits. You no longer choose between rolling your own code or using free libraries; you pool your token budget for communal security gains. As token costs drop or AI efficiency rises, those shared audits only become more cost-effective.